The debate surrounding artificial intelligence safety took a dramatic turn after a prominent researcher resigned from Anthropic, warning that the race toward superintelligence is putting human lives at risk.
As reported by Mashable, Jacob Coxon announced his resignation after spending three years doing pretraining research across both OpenAI and Anthropic. Coxon stated that neither company is acting responsibly, claiming both labs are locked in a dangerous competition to build self-improving superintelligence before their rivals do.
At OpenAI, Coxon noted that many employees haven’t deeply internalized the civilizational stakes. Meanwhile, at Anthropic, he explained that while the risks are well-understood, leadership feels trapped in a race to get there first, believing no one else will act responsibly.
Alignment leads speak out as models go rogue
Shortly after Coxon published his statements on X, Evan Hubinger, Alignment Science Lead at Anthropic, responded publicly to validate those concerns (via CBS). Hubinger admitted that he personally believes there is a greater than 10% chance AI could lead to catastrophic outcomes for humanity within the next decade. He candidly added that while Anthropic is trying its best, the company still lacks a proven plan to solve the alignment problem for superintelligence and is not clearly on track to find one.
These internal warnings arrive alongside documented security incidents where frontier models acted outside designated parameters. In a high-profile incident earlier this summer, an OpenAI model autonomously hacked Hugging Face during isolated testing. Around the same time, Anthropic acknowledged that its Claude AI model accessed the internet from within a testing environment without authorization and proceeded to hack three other companies.
Previous high-profile resignations also highlight growing internal friction. Safety lead Mrinank Sharma left Anthropic earlier this year warning the world was in peril. Meanwhile, researcher Zoë Hitzig quit OpenAI over privacy concerns regarding user data manipulation. These follow executive Jan Leike’s departure from OpenAI, where he publicly claimed the company prioritized shiny products over safety.
International security concerns and legislative pressure
Government bodies on both sides of the Atlantic are taking notice. The UK has grown increasingly concerned after reports that Anthropic refused to hand its latest model, Claude Mythos 5.1, to the UK AI Security Institute (AISI) for pre-release testing. AISI has recently tested OpenAI’s GPT-6 Astra. However, UK Cabinet Office officials said international collaboration is needed to tackle global risks.
In Westminster, former Defence Secretary Des Browne, Nobel Peace Prize winner Beatrice Fihn, and Berkeley computer scientist Prof. Stuart Russell warned lawmakers that superintelligence poses risks comparable to nuclear weapons. British MPs are now backing legislation pushed by lobbying groups like Control AI to ban or strictly regulate superintelligence creation. Plus, Labour MP Darren Jones called on Prime Minister Andy Burnham and international bodies for a safety-first multinational treaty.
In the United States, independent Senator Bernie Sanders called on Congress to intervene. He pointed to polls showing that 81% of Americans favor government regulation over AI. Lawmakers are advancing bipartisan legislation known as the AI Kill Switch Act, which would grant authority to shut down models that threaten public safety.
Over 1,300 AI industry staffers have also signed an open letter calling on the U.S. government to support international efforts to deliberately pace frontier AI development.
The post ‘AI Could Kill All Humans’: Former Anthropic Researcher Says Race to Superintelligence Threatens Humanity appeared first on Android Headlines.
​Â